[ODE] Angular-velocity autodisabling considered broken

[John Miles]

> I think that the problem that needs to be addressed is the jittering
> itself. Using the auto-disabling mechanism to reduce the effects may work
> well in some cases, but I think (or hope :) that other engines don't
> resort to such "hacks". Ideally, auto-disabling shouldn't affect the
> result of the simulation at all, only save some CPU time.
>

Ideally, yeah, but in reality I don't think you'll ever get rid of jitter,
at least not until we have perfect collision detection/response code that
has no epsilon bugs and can be allowed to generate a potentially-infinite
number of contacts.

As it stands now, jitter is a *huge* threat to overall system performance,
and it's fair game for heuristic solutions.

-- jm